Enterprise output value prediction method and device, computer device, readable storage medium and program product

Abstract

The application relates to an enterprise output value prediction method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: obtaining historical power consumption data of an enterprise to be predicted, performing clustering processing on the historical power consumption data to obtain power consumption mode features; obtaining enterprise data and regional macroeconomic data of the enterprise to be predicted, performing feature extraction on the historical power consumption data, the enterprise data and the regional macroeconomic data to obtain power consumption related features, macroeconomic features and time features, constructing an output value sample set based on the power consumption mode features, the power consumption related features, the macroeconomic features and the time features; selecting a target output value prediction model from a plurality of candidate output value prediction models based on the output value sample set; inputting an output value prediction sample of the enterprise to be predicted into a target output value prediction model corresponding to a relevant level of the output value prediction sample, and outputting an enterprise output value prediction value. The method can realize accurate prediction of enterprise output value.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for predicting enterprise output value. Background Technology

[0002] With the development of machine learning technology, the traditional method for forecasting the output value of companies in the automotive industry chain usually takes historical output value data as the core and uses time series models to extrapolate trends.

[0003] However, this method typically relies on the stationarity of the data and is poorly adaptable to external factors, making it difficult to capture the impact of non-stationary factors such as fluctuations in production load, changes in electricity consumption patterns, and adjustments in market policies on output value. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, device, computer equipment, computer-readable storage medium, and computer program product for enterprise output forecasting that can capture the impact of non-stationary factors such as production load fluctuations, changes in electricity consumption patterns, and market policy adjustments on output value, in order to address the above-mentioned technical problems.

[0005] Firstly, this application provides a method for predicting enterprise output value, including:

[0006] Obtain historical electricity consumption data of the enterprise to be predicted, perform clustering processing on the historical electricity consumption data, and obtain the electricity consumption pattern characteristics of the enterprise to be predicted.

[0007] Obtain enterprise data and regional macroeconomic data of the enterprises to be predicted, extract features from historical electricity consumption data, enterprise data and regional macroeconomic data to obtain electricity consumption-related features, macroeconomic features and time features, and construct an output value sample set based on electricity consumption pattern features, electricity consumption-related features, macroeconomic features and time features;

[0008] Based on the output value sample set, the target output value prediction model is selected from multiple candidate output value prediction models;

[0009] Obtain a sample of the output value forecast for the enterprise to be predicted. Based on the correlation coefficient between the output value forecast sample and the enterprise's output value, determine the corresponding correlation level. Input the output value forecast sample into the target output value forecast model corresponding to the correlation level, and output the enterprise output value forecast for the enterprise to be predicted.

[0010] In one embodiment, historical electricity consumption data includes point-by-point electricity consumption data for multiple production lines across multiple months; the historical electricity consumption data is clustered to obtain the electricity consumption pattern characteristics of the enterprise to be predicted, including:

[0011] For each month, based on the historical electricity consumption data for that month, the point-by-point electricity consumption data of each production line is converted into a two-dimensional daily electricity consumption feature matrix with date as the row and time node as the column;

[0012] A predetermined number of daily electricity consumption features are selected as initial cluster centers in the two-dimensional daily electricity consumption feature matrix;

[0013] For each initial cluster center, calculate the distance between each daily electricity consumption feature and the initial cluster center, and construct an electricity consumption pattern cluster based on the distance;

[0014] Calculate the target cluster center for the electricity consumption pattern cluster, determine whether the target cluster center meets the aggregation convergence condition, and if the target cluster center does not meet the aggregation convergence condition, replace the target cluster center with the initial cluster center and return to the step of calculating the distance between each daily electricity consumption feature and the initial cluster center to continue execution until the target cluster center meets the aggregation convergence condition.

[0015] Based on the multiple power consumption pattern clusters obtained when the target cluster centers satisfy the aggregation convergence condition, the power consumption pattern characteristics of multiple production lines in a month are determined.

[0016] In one embodiment, a sample set of output value is constructed based on electricity consumption pattern characteristics, electricity consumption-related characteristics, macroeconomic characteristics, and the time characteristics, including:

[0017] By splicing together the electricity consumption pattern characteristics, electricity-related characteristics, macroeconomic characteristics, and time characteristics of each month, the output value characteristics of multiple months can be obtained.

[0018] The output characteristics of each month are used as output samples to construct an output sample set.

[0019] In one embodiment, based on a sample set of output values, a target output value prediction model is selected from multiple candidate output value prediction models, including:

[0020] Based on the correlation coefficient between each output value sample and the enterprise output value in the output value sample set, the output value sample set is divided into multiple output value sample subsets under different correlation levels.

[0021] For each relevant level's output value sample subset, multiple initial output value prediction models to be trained are obtained. Based on the output value sample subset, each initial output value prediction model to be trained is trained to obtain multiple trained initial output value prediction models. Through particle swarm optimization algorithm, hyperparameters of each trained initial output value prediction model are optimized to obtain multiple candidate output value prediction models. Based on the mean absolute percentage error of each candidate output value prediction model, the target output value prediction model for the relevant level is selected from the multiple candidate output value prediction models.

[0022] In one embodiment, the correlation levels include a first correlation level, a second correlation level, and a third correlation level; the step of dividing the output value sample set into multiple output value sample subsets under multiple correlation levels based on the correlation coefficient between each output value sample and the enterprise's output value includes:

[0023] Calculate the correlation coefficient between each output value sample and the enterprise output value;

[0024] If the correlation coefficient is greater than the first preset correlation threshold, a subset of output value samples of the first correlation level is constructed based on the corresponding output value samples.

[0025] If the correlation coefficient is less than the first preset correlation threshold but greater than the second preset correlation threshold, a subset of output value samples of the second correlation level is constructed based on the corresponding output value samples.

[0026] If the correlation coefficient is less than the second preset correlation threshold but greater than the third preset correlation threshold, a subset of output value samples of the third correlation level is constructed based on the corresponding output value samples.

[0027] If the correlation coefficient is less than the third preset correlation threshold, the corresponding output value sample is removed from the output value sample set.

[0028] In one embodiment, the corresponding correlation level is determined based on the correlation coefficient between the output value forecast sample and the enterprise's output value, including:

[0029] Obtain the correlation coefficient calculation function;

[0030] Based on the output value forecast sample, the correlation coefficient between the output value forecast sample and the enterprise output value is obtained by solving the correlation coefficient calculation function.

[0031] Based on the relationship between the correlation coefficient of the output value forecast sample and the preset correlation threshold, the correlation level corresponding to the output value forecast sample is determined.

[0032] Secondly, this application also provides an enterprise output forecasting device, comprising:

[0033] The clustering module is used to obtain historical electricity consumption data of the enterprise to be predicted, and to perform clustering processing on the historical electricity consumption data to obtain the electricity consumption pattern characteristics of the enterprise to be predicted.

[0034] The module is used to acquire enterprise data and regional macroeconomic data of the enterprise to be predicted, extract features from historical electricity consumption data, enterprise data and regional macroeconomic data to obtain electricity consumption-related features, macroeconomic features and time features, and construct an output value sample set based on electricity consumption pattern features, electricity consumption-related features, macroeconomic features and the time features.

[0035] The filtering module is used to select the target output prediction model from multiple candidate output prediction models based on the output sample set;

[0036] The prediction module is used to obtain output value prediction samples of the enterprises to be predicted, determine the corresponding correlation level based on the correlation coefficient between the output value prediction samples and the enterprise output value, input the output value prediction samples into the target output value prediction model corresponding to the correlation level, and output the enterprise output value prediction value of the enterprises to be predicted.

[0037] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0038] Obtain historical electricity consumption data of the enterprise to be predicted, perform clustering processing on the historical electricity consumption data, and obtain the electricity consumption pattern characteristics of the enterprise to be predicted.

[0039] Obtain enterprise data and regional macroeconomic data of the enterprises to be predicted, extract features from historical electricity consumption data, enterprise data and regional macroeconomic data to obtain electricity consumption-related features, macroeconomic features and time features, and construct an output value sample set based on electricity consumption pattern features, electricity consumption-related features, macroeconomic features and time features;

[0040] Based on the output value sample set, the target output value prediction model is selected from multiple candidate output value prediction models;

[0041] Obtain a sample of the output value forecast for the enterprise to be predicted. Based on the correlation coefficient between the output value forecast sample and the enterprise's output value, determine the corresponding correlation level. Input the output value forecast sample into the target output value forecast model corresponding to the correlation level, and output the enterprise output value forecast for the enterprise to be predicted.

[0042] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0043] Obtain historical electricity consumption data of the enterprise to be predicted, perform clustering processing on the historical electricity consumption data, and obtain the electricity consumption pattern characteristics of the enterprise to be predicted.

[0044] Obtain enterprise data and regional macroeconomic data of the enterprises to be predicted, extract features from historical electricity consumption data, enterprise data and regional macroeconomic data to obtain electricity consumption-related features, macroeconomic features and time features, and construct an output value sample set based on electricity consumption pattern features, electricity consumption-related features, macroeconomic features and time features;

[0045] Based on the output value sample set, the target output value prediction model is selected from multiple candidate output value prediction models;

[0046] Obtain a sample of the output value forecast for the enterprise to be predicted. Based on the correlation coefficient between the output value forecast sample and the enterprise's output value, determine the corresponding correlation level. Input the output value forecast sample into the target output value forecast model corresponding to the correlation level, and output the enterprise output value forecast for the enterprise to be predicted.

[0047] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0048] Obtain historical electricity consumption data of the enterprise to be predicted, perform clustering processing on the historical electricity consumption data, and obtain the electricity consumption pattern characteristics of the enterprise to be predicted.

[0049] Obtain enterprise data and regional macroeconomic data of the enterprises to be predicted, extract features from historical electricity consumption data, enterprise data and regional macroeconomic data to obtain electricity consumption-related features, macroeconomic features and time features, and construct an output value sample set based on electricity consumption pattern features, electricity consumption-related features, macroeconomic features and time features;

[0050] Based on the output value sample set, the target output value prediction model is selected from multiple candidate output value prediction models;

[0051] Obtain a sample of the output value forecast for the enterprise to be predicted. Based on the correlation coefficient between the output value forecast sample and the enterprise's output value, determine the corresponding correlation level. Input the output value forecast sample into the target output value forecast model corresponding to the correlation level, and output the enterprise output value forecast for the enterprise to be predicted.

[0052] The aforementioned enterprise output forecasting methods, devices, computer equipment, computer-readable storage media, and computer program products acquire historical electricity consumption data of the enterprise to be forecasted, perform clustering processing on the historical electricity consumption data to obtain the electricity consumption pattern characteristics of the enterprise to be forecasted; acquire enterprise data and regional macroeconomic data of the enterprise to be forecasted, extract features from the historical electricity consumption data, enterprise data, and regional macroeconomic data to obtain electricity consumption-related features, macroeconomic features, and time features; and construct an output value sample set based on the electricity consumption pattern characteristics, electricity consumption-related features, macroeconomic features, and time features. Through the above process of constructing the output value sample and exploring the potential correlation between output value and various indicators, a comprehensive and stable feature input is formed, which can fully integrate multi-source information such as enterprise electricity consumption characteristics, electricity consumption pattern characteristics, macroeconomic indicators, and time features, optimizing the ability of multiple candidate output value forecasting models trained subsequently to capture output value changes. Based on the output value sample set, a target output value forecasting model is selected from multiple candidate output value forecasting models. Through the above process, the performance differences of different candidate output value prediction models in the context of automobile manufacturing enterprise data can be objectively evaluated, ensuring that the target output value prediction model is selected as the optimal solution for the current scenario, further improving prediction accuracy. The process involves obtaining output value prediction samples of the enterprise to be predicted, determining the corresponding correlation level based on the correlation coefficient between the output value prediction samples and the enterprise's output value, inputting the output value prediction samples into the target output value prediction model corresponding to the correlation level, and outputting the enterprise's output value prediction value. The enterprise output value prediction values ​​obtained in the above process can capture the impact of non-stationary factors such as production load fluctuations, changes in electricity consumption patterns, and market policy adjustments on output value. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a diagram illustrating the application environment of a business output forecasting method in one embodiment.

[0055] Figure 2 This is a flowchart illustrating a method for predicting enterprise output value in one embodiment;

[0056] Figure 3 This is a flowchart illustrating the enterprise output forecasting method in another embodiment;

[0057] Figure 4 This is a structural block diagram of an enterprise output forecasting device in one embodiment;

[0058] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0060] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0061] The enterprise output forecasting method provided in this application embodiment can be applied to, for example, Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104, or it can be located in the cloud or on another network server. Specifically, terminal 102 or server 104 executes an enterprise output forecasting method, which includes: acquiring historical electricity consumption data of the enterprise to be forecasted; clustering the historical electricity consumption data to obtain the electricity consumption pattern characteristics of the enterprise to be forecasted; acquiring enterprise data and regional macroeconomic data of the enterprise to be forecasted; extracting features from the historical electricity consumption data, enterprise data, and the regional macroeconomic data to obtain electricity consumption-related features, macroeconomic features, and time features; constructing an output value sample set based on the electricity consumption pattern characteristics, electricity consumption-related features, macroeconomic features, and time features; selecting a target output value forecasting model from multiple candidate output value forecasting models based on the output value sample set; acquiring output value forecasting samples of the enterprise to be forecasted; determining the corresponding correlation level based on the correlation coefficient between the output value forecasting samples and the enterprise output value; inputting the output value forecasting samples into the target output value forecasting model corresponding to the correlation level; and outputting the enterprise output value forecasting value of the enterprise to be forecasted. Terminal 102 can be, but is not limited to, various personal computers, laptops, etc. Server 104 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0062] In one exemplary embodiment, such as Figure 2 As shown, a method for predicting enterprise output value is provided, and this method is applied to... Figure 1 Taking terminal 102 as an example, the explanation includes the following steps 202 to 208. Wherein:

[0063] Step 202: Obtain historical electricity consumption data of the enterprise to be predicted, perform clustering processing on the historical electricity consumption data, and obtain the electricity consumption pattern characteristics of the enterprise to be predicted.

[0064] The companies to be predicted can be enterprises in the automotive industry chain, whose production lines can include passenger car production lines, multi-purpose vehicle (MPV) production lines, new energy vehicle production lines, and sport utility vehicle (SUV) production lines. Historical electricity consumption data includes point-by-point electricity consumption data for each time point within the target historical period. Electricity consumption pattern characteristics refer to the number of high-load days, medium-load days, and low-load days of the production line within a month.

[0065] Optionally, the target historical time period refers to any time period before the current time point, including multiple months. Multiple historical months can be selected as the target historical time period. For example, 36 months before the current time point can be selected as the target historical time period.

[0066] For example, a categorized time-series repair strategy is adopted to address three types of anomalies in historical electricity consumption data: negative values, missing values, and cumulative value regression. Specifically, a time-series forward imputation method is used for numerical anomalies and missing data; for cumulative electricity regression anomalies, a rule is constructed to determine the proportion of the historical maximum value. If the ratio of the historical maximum value to the current value is less than 2, it is determined to be a data statistical error, and the current value is directly replaced with the historical maximum value; if it is greater than or equal to 2, it indicates that the meter has been reset to zero, and the sum of the historical maximum value and the current value is used to replace the current value. After the above processing, the preprocessed historical electricity consumption data is obtained.

[0067] For example, the preprocessed historical electricity consumption data is divided according to the production line dimension to obtain the production line electricity consumption data for each production line. For the production line electricity consumption data of each production line, daily electricity consumption features are extracted to obtain multiple daily electricity consumption features of the production line. The multiple daily electricity consumption features are then clustered to obtain the electricity consumption pattern features of the production line.

[0068] Step 204: Obtain enterprise data and regional macroeconomic data of the enterprise to be predicted. Extract features from historical electricity consumption data, enterprise data and regional macroeconomic data to obtain electricity consumption-related features, macroeconomic features and time features. Based on electricity consumption pattern features, electricity consumption-related features, macroeconomic features and time features, construct an output value sample set.

[0069] The enterprise data includes data related to the enterprise to be predicted, such as user ID, unit name, measurement point number, and historical output value. The regional macroeconomic data includes economic data for the region where the enterprise to be predicted is located, including total industrial output of enterprises above a designated size, total retail sales of consumer goods, port cargo throughput, container throughput, total import and export value, total import value, total export value, secondary industry output, light industry, heavy industry, equipment manufacturing, automobile manufacturing, auto parts manufacturing, chemical manufacturing, high-tech industry output, export delivery value, and primary forms of production such as plastics, steel, automobiles, and automobile engines.

[0070] For example, abnormally negative values ​​in numerical indicators (such as primary plastic production) in enterprise data and regional macroeconomic data are corrected using time series interpolation. The specific formula used in time series interpolation is: Corrected value = Original value of the previous month + Forecast difference value. For missing monthly data, it is supplemented by combining the proportion of working days and cumulative values. For indicators with excessively large missing values ​​and extremely weak correlations, such as monthly data from the chemical manufacturing industry, they are deleted. After the above processing, the preprocessed enterprise data and regional macroeconomic data are obtained.

[0071] For example, the electricity consumption data of each production line is summarized and calculated on a monthly basis, and the monthly electricity consumption of each production line is added together to obtain the monthly electricity consumption data of the enterprise to be predicted.

[0072] For example, feature extraction is performed on historical electricity consumption data to obtain electricity-related features, and feature extraction is performed on enterprise data and regional macroeconomic data to obtain macroeconomic features and time features. The electricity consumption pattern features, electricity consumption-related features, macroeconomic features, and time features are then concatenated to obtain multiple output value samples, and an output value sample set is constructed based on these multiple output value samples.

[0073] Step 206: Based on the output value sample set, select the target output value prediction model from multiple candidate output value prediction models.

[0074] Among them, the candidate output prediction model is an output prediction model trained using an output sample set. The target output prediction model is the optimal output prediction model that best suits the characteristics of the enterprise's data.

[0075] Alternatively, a pre-trained random forest model, a pre-trained XG Boost model, or a pre-trained SARIMAX model can be used as a candidate output prediction model.

[0076] For example, multiple initial output prediction models to be trained are obtained, and each initial output prediction model is trained using an output sample set to obtain an initial output prediction model. Hyperparameter optimization is then performed on each initial output prediction model to obtain multiple candidate output prediction models.

[0077] Optionally, the target output prediction model is selected from multiple candidate output prediction models based on the average absolute percentage error of each candidate output prediction model.

[0078] Step 208: Obtain the output value prediction sample of the enterprise to be predicted. Based on the correlation coefficient between the output value prediction sample and the enterprise's output value, determine the corresponding correlation level. Input the output value prediction sample into the target output value prediction model corresponding to the correlation level, and output the enterprise output value prediction value of the enterprise to be predicted.

[0079] Among them, the output value forecast sample is the output value sample used to predict the output value of the enterprise to be predicted at the current time point.

[0080] For example, the correlation coefficient between the output value forecast sample and the enterprise output value is calculated using the correlation coefficient calculation function, the correlation level corresponding to the correlation coefficient is determined, and this correlation level is used as the correlation level of the output value forecast sample.

[0081] The aforementioned enterprise output forecasting method involves acquiring historical electricity consumption data of the enterprise to be forecasted, clustering this data to obtain the enterprise's electricity consumption pattern characteristics, acquiring enterprise data and regional macroeconomic data, and extracting features from these data to obtain electricity consumption-related features, macroeconomic features, and time features. Based on these features, an output value sample set is constructed. This process of constructing the output value sample and exploring the potential correlation between output value and various indicators forms a comprehensive and stable feature input. It fully integrates multi-source information such as enterprise electricity consumption characteristics, electricity consumption pattern characteristics, macroeconomic indicators, and time features, optimizing the ability of multiple candidate output value forecasting models trained subsequently to capture output value changes. Based on the output value sample set, a target output value forecasting model is selected from multiple candidate models. This process objectively evaluates the performance differences of different candidate output value forecasting models in the context of automobile manufacturing enterprise data, ensuring that the selected target output value forecasting model is the optimal solution for the current scenario, further improving forecasting accuracy. The process involves obtaining a sample of output value forecasts for the enterprises to be predicted. Based on the correlation coefficient between the output value forecast samples and the enterprise output value, the corresponding correlation level is determined. The output value forecast samples are then input into the target output value forecast model corresponding to the correlation level, and the output value forecast for the enterprise to be predicted is output. The enterprise output value forecasts obtained in the above process can capture the impact of non-stationary factors such as production load fluctuations, changes in electricity consumption patterns, and market policy adjustments on output value.

[0082] In an exemplary embodiment, historical electricity consumption data includes point-by-point electricity consumption data for multiple production lines across multiple months. Clustering the historical electricity consumption data to obtain the electricity consumption pattern characteristics of the enterprise to be predicted includes: for each month, based on the historical electricity consumption data for that month, converting the point-by-point electricity consumption data of each production line into a two-dimensional daily electricity consumption feature matrix with dates as rows and time nodes as columns; selecting a preset number of daily electricity consumption features as initial cluster centers in the two-dimensional daily electricity consumption feature matrix; for each initial cluster center, calculating the distance between each daily electricity consumption feature and the initial cluster center, and constructing an electricity consumption pattern cluster based on the distance; calculating the target cluster center for the electricity consumption pattern cluster, determining whether the target cluster center meets the aggregation convergence condition, and if the target cluster center does not meet the aggregation convergence condition, replacing the target cluster center with the initial cluster center, and returning to the step of calculating the distance between each daily electricity consumption feature and the initial cluster center to continue execution until the target cluster center meets the aggregation convergence condition; and determining the electricity consumption pattern characteristics of multiple production lines in the month based on the multiple electricity consumption pattern clusters obtained when the target cluster center meets the aggregation convergence condition.

[0083] The two-dimensional daily electricity consumption feature matrix includes multiple daily electricity consumption features. Electricity consumption pattern features refer to the number of high-load days, medium-load days, and low-load days for the enterprise as a whole and for each production line, as well as the relative output of each production line under different electricity consumption patterns. Electricity consumption patterns include high load, medium load, and low load. The aggregation convergence condition is that the target cluster centers no longer change significantly.

[0084] Optionally, the preset quantity is a pre-set value that is related to the number of production lines of the enterprise to be predicted. It is generally three times the number of production lines of the enterprise to be predicted. For example, if the preset quantity is k, then k daily electricity consumption characteristics will be randomly selected as the initial cluster centers.

[0085] Optionally, before selecting a preset number of daily electricity consumption features as initial cluster centers, L2 norm normalization is performed on the two-dimensional daily electricity consumption feature matrix to eliminate the dimensional differences in electricity consumption at different time points, so as to keep the contribution of the data dimension of each daily electricity consumption feature to the clustering result balanced.

[0086] For example, for each initial cluster center, the distance between each daily electricity consumption feature and the initial cluster center is calculated, and each daily electricity consumption feature is assigned to the vicinity of the nearest initial cluster center to construct a preset number of electricity consumption pattern clusters.

[0087] Alternatively, the distance can be calculated using Euclidean distance.

[0088] For example, for each electricity consumption pattern cluster obtained when the target cluster center satisfies the aggregation convergence condition, the number of daily electricity consumption features existing in the electricity consumption pattern cluster is calculated, and the production line corresponding to the electricity consumption pattern cluster is determined. The number of existing daily electricity consumption features is used as the electricity consumption pattern feature of the production line corresponding to the electricity consumption pattern cluster in the month.

[0089] Furthermore, for each production line, three power consumption mode clusters corresponding to that production line are obtained, namely the high load cluster, medium load cluster, and low load cluster of the production line. Based on the high load cluster, medium load cluster, and low load cluster of the production line, the power consumption mode characteristics of the production line are determined.

[0090] In this embodiment, by clustering and classifying daily electricity consumption data, the production load distribution patterns of different production lines in each month can be accurately identified. More distinctive production pattern information can be extracted from the electricity consumption time series data, avoiding the obscuring of the true production patterns due to fluctuations in electricity consumption data, and providing a feature basis that is more in line with the actual production status of enterprises for subsequent output value prediction.

[0091] In one embodiment, a value sample set is constructed based on electricity consumption pattern characteristics, electricity consumption-related characteristics, macroeconomic characteristics, and the time characteristics, including: concatenating the electricity consumption pattern characteristics, electricity consumption-related characteristics, macroeconomic characteristics, and time characteristics of each month to obtain value characteristics for multiple months; and using the value characteristics of each month as value samples to construct the value sample set.

[0092] The electricity consumption pattern characteristics include the number of high-load days, medium-load days, and low-load days for each production line (sedans, MPVs, new energy vehicles, SUVs) and the enterprise as a whole; and the relative output of each production line under different electricity consumption patterns. Relative output is the product of the average electricity consumption under different electricity consumption patterns and the corresponding number of load days. Macroeconomic characteristics include regional macroeconomic indicators such as the total industrial output of enterprises above designated size, total retail sales of consumer goods, port cargo throughput, container throughput, total import and export value, total import value, total export value, secondary industry output (quarterly), light industry, heavy industry, equipment manufacturing, automobile manufacturing, auto parts manufacturing, high-tech industry output, export delivery value, primary forms of plastics, steel, automobiles, automobile engines, and the total industrial output of enterprises above designated size (quarterly), covering dimensions such as industrial production, foreign trade, and industrial structure. Time characteristics include year, month, and quarter.

[0093] Furthermore, electricity consumption-related characteristics include basic electricity consumption characteristics, statistical electricity consumption characteristics, ratio-based electricity consumption characteristics, trend characteristics, and industry-related characteristics. Basic electricity consumption characteristics include the enterprise's electricity consumption this month, electricity consumption 1-3 months prior, and cumulative electricity consumption for the current year; statistical electricity consumption characteristics include the median, average, maximum, minimum, and standard deviation of electricity consumption over the past 3 / 6 / 9 months; ratio-based electricity consumption characteristics include the ratio of this month's electricity consumption to the statistical values ​​(median, average, etc.) of electricity consumption over the past 3 / 6 / 9 months, and the ratio of this month's electricity consumption to the statistical values ​​of electricity consumption for the current year; trend characteristics include the month-on-month and year-on-year growth rates of electricity consumption; and industry-related characteristics include the average selling price of each vehicle model.

[0094] For example, the electricity consumption pattern characteristics, electricity consumption-related characteristics, macroeconomic characteristics, and time characteristics of each month are concatenated to obtain output value characteristics for multiple months. Each output value characteristic is used as an output value sample, resulting in multiple output value samples. An output value sample set is then constructed based on these multiple output value samples.

[0095] In this embodiment, by fusing and splicing multi-dimensional features, the system fully integrates multi-source information such as the enterprise's own production electricity consumption patterns, regional macroeconomic trends, and time cycle changes. This avoids prediction bias caused by single feature inputs and provides training samples with complete information dimensions that fit the actual production and operation situation for subsequent model training. This helps the model better learn the potential correlation between different features and enterprise output value, thereby improving the accuracy of the final prediction results.

[0096] In one embodiment, selecting a target output value prediction model from multiple candidate output value prediction models based on an output value sample set includes: dividing the output value sample set into multiple output value sample subsets under multiple correlation levels according to the correlation coefficient between each output value sample and the enterprise's output value; obtaining multiple initial output value prediction models to be trained for each output value sample subset under each correlation level; training each initial output value prediction model to be trained based on the output value sample subset to obtain multiple trained initial output value prediction models; optimizing the hyperparameters of each trained initial output value prediction model using a particle swarm optimization algorithm to obtain multiple candidate output value prediction models; and selecting a target output value prediction model of a correlation level from the multiple candidate output value prediction models based on the mean absolute percentage error of each candidate output value prediction model.

[0097] The enterprise output value is historical enterprise output data. Relevance levels include first, second, and third correlation levels. The output value sample subset includes at least one output value sample. Mean Absolute Percentage Error (MAPE) is a commonly used metric for evaluating the accuracy of predictive models. Particle Swarm Optimization (PSO) is a swarm intelligence-based optimization algorithm that simulates the foraging behavior of flocks of birds or schools of fish, finding the optimal solution by updating the position and velocity of particles in the solution space.

[0098] For example, the output sample subset is divided into training, validation, and test sets according to time. The test set can be output samples from a period relatively far removed from the current time point, such as output samples from the previous 5 to 4 years. The validation and test sets can be output samples from a period relatively close to the current time point, such as output samples from the previous two years. The training set is standardized using a standard scaler. The standard scaler is a numerical feature standardization method, and its core formula is... ,in It is the mean of the features. It is the standard deviation of the feature. It scales the numerical feature to a distribution with a mean of 0 and a standard deviation of 1.

[0099] Optionally, the formula for calculating the mean absolute percentage error is:

[0100]

[0101] in, It is the actual value. It is a predicted value. It represents the number of output value samples.

[0102] Optionally, a random forest model can be chosen as the initial output prediction model to be trained. The random forest model constructs multiple independent decision trees, outputs the prediction results of each decision tree, and calculates the average of the prediction results as the final prediction value. This can be represented as:

[0103]

[0104] in, For the test sample feature vector, For the first The predicted value of each decision tree for this sample. This represents the total number of decision trees.

[0105] Furthermore, when the initial output prediction model to be trained is a random forest model, the process of training the initial output prediction model to obtain a trained initial output prediction model based on a subset of output samples is as follows: For a subset of output samples containing n output samples and M features, n output samples are randomly selected with replacement to generate the training set of the k-th tree. The output samples that are not selected can be used for model evaluation when there is no validation set. When splitting at each node of the k-th tree, m features are randomly selected from the M features, and the optimal split point is calculated based only on this subset of output samples. The split is performed based on the optimal split. A single decision tree is constructed using the Classification and Regression Tree (CART) algorithm, and the single tree is not pruned. The above steps are repeated to generate K independent decision trees (K is the number of trees), and all trees have no weight difference, thus obtaining the trained initial output prediction model.

[0106] Optionally, the XG Boost model is selected as the initial output prediction model to be trained. The XG Boost model is an optimization ensemble learning algorithm based on the Gradient Boosting Decision Tree (GBDT) framework. It constructs multiple decision trees and outputs the prediction results in a weighted summation manner. This can be represented as:

[0107]

[0108] in, Let i be the i-th output characteristic; Let K be the predicted value of the i-th output feature by the k-th decision tree, where K is the total number of decision trees. This is the predicted result.

[0109] Furthermore, when the initial output prediction model to be trained is an XG Boost model, the process of training the initial output prediction model to obtain a trained initial output prediction model based on a subset of output samples includes the following steps: selecting a constant term as the initial prediction value, which can typically be the mean of the enterprise's output prediction values, i.e. ,in, Let be the loss function; calculate the gradient, that is, in the t-th iteration, calculate the negative gradient of the loss function with respect to the current model's predictions, and use it as the training objective for the new tree, i.e.: The t-th regression tree is trained using the CART algorithm, with the goal of fitting the residuals. Simultaneously, regularization constraints are introduced to limit the complexity of the tree (such as limiting tree depth and the number of leaf nodes, and adding L2 regularization to penalize leaf node weights). An optimal split point is found using a greedy algorithm. The weights of the leaf nodes in the t-th tree are optimized using a second-order Taylor expansion to balance the fitting effect and model complexity. The specific formula is as follows: ,in, For the first-order gradient, It is a second-order gradient. The L2 regularization coefficient, Given the output value sample of the j-th leaf node; weight and accumulate the predicted values ​​of the t-th tree into the existing model to update the prediction results: ,in, Set the learning rate; repeat the above steps until the preset number of trees K is reached or the loss function converges. Use the final model as the trained initial output prediction model.

[0110] Optionally, the SARIMAX model is selected as the initial output prediction model to be trained. It is an extended form of the SARIMAX model, and its core structure can be represented as:

[0111]

[0112] in, This is the non-seasonal part, including p for the autoregressive (AR) phase, d for the difference order, and q for the moving average; The seasonal component includes P, which is the seasonal autoregressive order; D, which is the seasonal difference order; Q, which is the seasonal moving average order; and s, which is the seasonal cycle length (e.g., daily cycle s=24, monthly cycle s=12). These refer to external explanatory variables that affect the target variable.

[0113] Furthermore, the firm output value predicted by the SARIMAX model is:

[0114]

[0115] in, It is a stationary series after d non-seasonal differencing and D seasonal differencing; , , The non-seasonal AR coefficient, , The MA coefficient is non-seasonal. , , The seasonal AR coefficient, , This is the seasonal SMA coefficient; Let k be exogenous variables at time t. The regression coefficients are for exogenous variables; For random error term, is the series mean, and s is the seasonal cycle length.

[0116] Furthermore, in the final prediction, it is necessary to use inverse difference operations to... Predicted values ​​restored to the original sequence When the initial output prediction model to be trained is a SARIMAX model, the process of training the initial output prediction model based on a subset of output samples to obtain a trained initial output prediction model includes the following steps: Perform a stationarity test on each output sample in the subset of output samples; if it is not stationary, it needs to be transformed into a stationary sequence through differencing; screen exogenous variables that are strongly correlated with the target variable, such as through correlation analysis and feature importance screening, and perform preprocessing operations such as standardization / normalization and missing value imputation on the exogenous variables; determine the model parameters (p,d,q)s using visualization tools (autocorrelation function ACF plot, partial autocorrelation function PACF plot), and remove redundant exogenous variables through correlation analysis, stepwise regression, or regularization methods, retaining variables that have significant explanatory power for the target variable; construct the SARIMAX model based on the time-series parameters after order determination and the screened exogenous variables, and simultaneously solve the time-series coefficients using maximum likelihood estimation. ) and regression coefficients of exogenous variables ( At the same time, it is checked whether the model residuals are white noise. If the residuals have significant autocorrelation, it indicates that the time series parameters have not fully captured the features and need to be reordered. If the residuals still have systematic bias, the selection of exogenous variables needs to be optimized. Finally, the SARIMAX model with the smallest MAPE is selected as the trained initial output prediction model.

[0117] For example, the particle swarm optimization algorithm uses the mean absolute percentage error (MAPE) calculated from the performance of the trained initial output prediction model on the validation set as the optimization objective, and optimizes the key hyperparameters of each trained initial output prediction model. The specific steps of the particle swarm optimization algorithm include:

[0118] Step 1: Initialize the population for optimizing parameters.

[0119] First, it is necessary to determine the search range (lower bound) of the optimization parameters. Upper Realm Secondly, N particles are randomly generated within the search range. The number of particles is related to the population size, and the initial position of each particle is... Uniform sampling within the parameter range, initial velocity Random generation is typically within the range of [-1, 1] or a parameter range ratio; finally, the individual optimal position of each particle is initialized. Calculate the initial fitness value And determine the global optimal position of the population. .

[0120] Step 2: Update the particles through iteration.

[0121] First, based on inertia weights, individual cognition, and group cooperation, the initial velocity is... Adjustments will be made, and the specific formula is as follows:

[0122]

[0123] in, The inertia weight can be set to 0.8; , This is the learning factor, typically set to 0.5; The value is a random number in the range [0,1]. Next, the particle's position is updated based on the new velocity, using the following formula: The updated particle position must be within the upper and lower bounds of the parameter. To avoid parameters exceeding the valid range, the fitness value at the new position is calculated based on the updated position of the particle. Based on adaptation to the new location, update If a certain particle's Better than the current Then update .

[0124] Step 3: Determine whether the iteration termination condition is met.

[0125] The iteration termination condition is reaching the preset maximum number of iterations, or the global optimal fitness value not changing significantly over multiple consecutive generations, for example, the change is less than [a certain value]. If the iteration termination condition is met, terminate the iteration; if the iteration termination condition is not met, return to step 2 and continue execution.

[0126] Step 4: Find the globally optimal position corresponding to the iteration termination condition. The optimal combination of key hyperparameters is used as the basis for the final training of the model.

[0127] In this embodiment, multiple initial output prediction models are trained to obtain multiple well-trained initial output prediction models. In addition, the hyperparameters of different initial output prediction models are optimized by using the particle swarm optimization algorithm, which can effectively balance the model's fitting ability and generalization performance, avoid the subjectivity and inefficiency of manual parameter tuning, and improve the accuracy of the final prediction results.

[0128] In an exemplary embodiment, the correlation levels include a first correlation level, a second correlation level, and a third correlation level. The step of dividing the output value sample set into multiple output value sample subsets under multiple correlation levels based on the correlation coefficient between each output value sample and the enterprise's output value includes: calculating the correlation coefficient between each output value sample and the enterprise's output value; constructing a first correlation level output value sample subset based on the corresponding output value sample when the correlation coefficient is greater than a first preset correlation threshold; constructing a second correlation level output value sample subset based on the corresponding output value sample when the correlation coefficient is less than the first preset correlation threshold but greater than a second preset correlation threshold; constructing a third correlation level output value sample subset based on the corresponding output value sample when the correlation coefficient is less than the third preset correlation threshold; and removing the corresponding output value sample from the output value sample set when the correlation coefficient is less than the third preset correlation threshold.

[0129] Among them, the first correlation level, the second correlation level, and the third correlation level represent strong correlation, moderate correlation, and weak correlation, respectively.

[0130] Optionally, the correlation coefficient can be a Pearson correlation coefficient, with the absolute value ranging from [0,1]. The correlation coefficient between each output sample and the enterprise's output is calculated using the Pearson coefficient calculation method. A higher correlation coefficient indicates a stronger correlation between the output sample and the enterprise's output. The first preset correlation threshold is 0.6, the second preset correlation threshold is 0.4, and the third preset correlation threshold is 0.2.

[0131] In this embodiment, by dividing output samples with different correlations into subsets of output samples with corresponding correlation levels, features with weak correlation but still predictive value are retained, while almost unrelated noisy features are eliminated. This can reduce the model input dimension, reduce redundant calculations, and improve training efficiency while ensuring the amount of feature information. It can also select the optimal model based on the correlation of the output prediction samples when using output prediction samples for subsequent output prediction.

[0132] In one embodiment, determining the corresponding correlation level based on the correlation coefficient between the output value prediction sample and the enterprise's output value includes: obtaining a correlation coefficient calculation function; obtaining the correlation coefficient between the output value prediction sample and the enterprise's output value by solving the correlation coefficient calculation function based on the output value prediction sample; and determining the correlation level corresponding to the output value prediction sample based on the relationship between the correlation coefficient of the output value prediction sample and a preset correlation threshold.

[0133] The preset related thresholds include a first preset related threshold, a second preset related threshold, and a third preset related threshold.

[0134] Alternatively, the correlation coefficient calculation function can be the Pearson correlation coefficient calculation function:

[0135]

[0136] in, The correlation coefficient is... The mean of the output value forecast sample. This represents the average output value of the enterprise. For the first sample of output value forecast One data point, For the first The output value of each enterprise.

[0137] For example, the correlation coefficient is compared with the first preset correlation threshold, the second preset correlation threshold and the third preset correlation threshold in the preset correlation threshold to determine the correlation level corresponding to the output value prediction sample.

[0138] In this embodiment, the correlation level of the output value prediction sample is determined by the correlation coefficient, which can quickly match the target output value prediction model trained with the corresponding correlation level. There is no need to call all models for prediction, which reduces unnecessary calculations while ensuring prediction accuracy and improving the efficiency of output value prediction.

[0139] In one of the most specific embodiments, such as Figure 3 As shown, the following section will use a car manufacturing company as an example to introduce a method for predicting enterprise output value according to this application.

[0140] First, multi-source data collection and data preprocessing are performed.

[0141] It is necessary to collect historical electricity consumption data of the automobile manufacturer, perform clustering processing on the historical electricity consumption data to obtain the electricity consumption pattern characteristics of the company to be predicted; at the same time, collect the company's enterprise data and regional macroeconomic data, and extract features from the historical electricity consumption data, enterprise data and regional macroeconomic data to obtain electricity consumption-related features, macroeconomic features and time features.

[0142] Secondly, construct a sample set of output value.

[0143] The electricity consumption pattern characteristics, electricity-related characteristics, macroeconomic characteristics, and time characteristics of each month are spliced ​​together to obtain the output value characteristics of multiple months; the output value characteristics of each month are used as output value samples to construct an output value sample set.

[0144] Next, based on the output value sample set, the target output value prediction model is selected from multiple candidate output value prediction models.

[0145] Based on the correlation coefficients between each output value sample and the enterprise output value in the output value sample set, the output value sample set is divided into multiple output value sample subsets under different correlation levels. For each output value sample subset under a correlation level, multiple initial output value prediction models to be trained are obtained. Based on the output value sample subsets, each initial output value prediction model to be trained is trained to obtain multiple trained initial output value prediction models. Through particle swarm optimization algorithm, hyperparameter optimization is performed on each trained initial output value prediction model to obtain multiple candidate output value prediction models. Based on the mean absolute percentage error of each candidate output value prediction model, the target output value prediction model of the correlation level is selected from the multiple candidate output value prediction models.

[0146] Enterprise output value is based on historical enterprise output data. The correlation levels are categorized as Level 1, Level 2, and Level 3.

[0147] Based on the correlation coefficients between each output value sample and the enterprise output value in the output value sample set, the output value sample set is divided into multiple output value sample subsets under different correlation levels. For each output value sample subset under a different correlation level, initial random forest models, initial XG Boost models, and initial SARIMAX models are obtained for training. Based on the output value sample subsets, the initial random forest models, initial XG Boost models, and initial SARIMAX models are trained to obtain trained initial random forest models, initial XG Boost models, and initial SARIMAX models. Hyperparameter optimization is performed on the trained initial random forest models, initial XG Boost models, and initial SARIMAX models using the particle swarm optimization algorithm to obtain candidate output value prediction models, which are the random forest model, the XG Boost model, and the SARIMAX model.

[0148] The optimal parameter combination for the first-level random forest model is 500 trees, a maximum tree depth of 12, a minimum number of samples required for splitting of 10, a minimum number of samples per leaf node of 1, and a sqrt number of features used in each split. The optimal parameter combination for the second-level random forest model is 266 trees, a maximum tree depth of 15, a minimum number of samples required for splitting of 10, a minimum number of samples per leaf node of 1, and a sqrt number of features used in each split. The optimal parameter combination for the third-level random forest model is 50 trees, a maximum tree depth of 15, a minimum number of samples required for splitting of 11, a minimum number of samples per leaf node of 5, and a sqrt number of features used in each split.

[0149] The optimal parameter combination for the XG Boost model at the first relevance level is: 100 trees, maximum tree depth of 10, learning rate of 0.0895, sample sampling ratio of 0.6, feature sampling ratio of 1.0, minimum loss reduction from splits of 1.0, and minimum sample weight of child nodes of 10. The optimal parameter combination for the XG Boost model at the second relevance level is: 181 trees, maximum tree depth of 4, learning rate of 0.0549, sample sampling ratio of 0.9719, feature sampling ratio of 0.9831, minimum loss reduction from splits of 0.6430, and minimum sample weight of child nodes of 8. The optimal parameter combination for the XG Boost model at the third relevance level is: 356 trees, maximum tree depth of 3, learning rate of 0.0740, sample sampling ratio of 0.9260, feature sampling ratio of 0.7712, minimum loss reduction from splits of 0.4377, and minimum sample weight of child nodes of 9.

[0150] The optimal parameter combination for the SARIMAX model at the first correlation level is: P (autoregressive order) 1, d (difference order) 0, q (moving average order) 0, P (seasonal autoregressive order) 0, D (seasonal difference order) 0, seasonal moving average order 0, s (seasonal cycle) 12; The optimal parameter combination for the SARIMAX model at the second correlation level is: P (autoregressive order) 2, d (difference order) 1, q (moving average order) 1, P (seasonal autoregressive order) 1, D (seasonal difference order) 0, seasonal moving average order 0, s (seasonal cycle) 12; The optimal parameter combination for the SARIMAX model at the third correlation level is: P (autoregressive order) 1, d (difference order) 1, q (moving average order) 1, P (seasonal autoregressive order) 1, D (seasonal difference order) 1, seasonal moving average order 0, s (seasonal cycle) 12.

[0151] Based on the optimal parameter combination for each model, the model is trained and used as the final prediction model. The MAPE results on the test set are shown in Table 1.

[0152] Table 1

[0153] Final prediction model First preset relevant threshold Second preset related threshold Third preset related threshold Random Forest Model 6.11% 8.20% 7.72% XG Boost model 5.13% 2.36% 2.10% SARIMAX model 3.69% 7.71% 8.58%

[0154] The MAPE values ​​of the Random Forest model range from 6.11% to 8.20%, the XG Boost model from 2.36% to 5.13%, and the SARIMAX model from 3.69% to 8.58%. According to Table 1, a lower MAPE indicates higher prediction accuracy for the candidate output prediction models. Therefore, the SARIMAX model is selected as the target output prediction model for the first correlation level; the XG Boost model is selected as the target output prediction model for the second and third correlation levels.

[0155] Finally, enterprise output forecast.

[0156] Obtain the enterprise's output value forecast sample, determine the corresponding correlation level based on the correlation coefficient between the output value forecast sample and the enterprise's output value, input the output value forecast sample into the target output value forecast model corresponding to the correlation level, and output the enterprise's output value forecast value.

[0157] Assuming the company's relevance level is level three, and XG Boost achieves the best prediction accuracy with a mean MAPE of 2.10% at the third preset relevance threshold, then selecting the XG Boost model as the target output prediction model can meet the company's needs for accurate output prediction.

[0158] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0159] Based on the same inventive concept, this application also provides an enterprise output forecasting device for implementing the enterprise output forecasting method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more enterprise output forecasting device embodiments provided below can be found in the limitations of the enterprise output forecasting method described above, and will not be repeated here.

[0160] In one exemplary embodiment, such as Figure 4 As shown, an enterprise output forecasting device 400 is provided, including: a clustering module 402, a construction module 404, a screening module 406, and a forecasting module 408, wherein:

[0161] Clustering module 402 is used to obtain historical electricity consumption data of the enterprise to be predicted, perform clustering processing on the historical electricity consumption data, and obtain the electricity consumption pattern characteristics of the enterprise to be predicted.

[0162] The construction module 404 is used to acquire enterprise data and regional macroeconomic data of the enterprise to be predicted, extract features from historical electricity consumption data, enterprise data and regional macroeconomic data to obtain electricity consumption-related features, macroeconomic features and time features, and construct an output value sample set based on electricity consumption pattern features, electricity consumption-related features, macroeconomic features and the time features.

[0163] The screening module 406 is used to select the target output prediction model from multiple candidate output prediction models based on the output sample set.

[0164] The prediction module 408 is used to obtain the output value prediction sample of the enterprise to be predicted, determine the corresponding correlation level based on the correlation coefficient between the output value prediction sample and the enterprise output value, input the output value prediction sample into the target output value prediction model corresponding to the correlation level, and output the enterprise output value prediction value of the enterprise to be predicted.

[0165] In one embodiment, the clustering module 402 is further configured to, for each month, convert the point-by-point electricity consumption data of each production line into a two-dimensional daily electricity consumption feature matrix with dates as rows and time nodes as columns, based on the historical electricity consumption data of the month; select a preset number of daily electricity consumption features as initial cluster centers in the two-dimensional daily electricity consumption feature matrix; for each initial cluster center, calculate the distance between each daily electricity consumption feature and the initial cluster center, and construct an electricity consumption pattern cluster based on the distance; calculate the target cluster center of the electricity consumption pattern cluster, determine whether the target cluster center meets the aggregation convergence condition, and if the target cluster center does not meet the aggregation convergence condition, replace the target cluster center with the initial cluster center, and return to the step of calculating the distance between each daily electricity consumption feature and the initial cluster center to continue execution until the target cluster center meets the aggregation convergence condition; and determine the electricity consumption pattern features of multiple production lines in the month based on the multiple electricity consumption pattern clusters obtained when the target cluster center meets the aggregation convergence condition.

[0166] In one embodiment, the construction module 404 is further configured to concatenate the electricity consumption pattern characteristics, electricity consumption-related characteristics, macroeconomic characteristics and time characteristics of each month to obtain the output value characteristics of multiple months; and use the output value characteristics of each month as output value samples to construct an output value sample set.

[0167] In one embodiment, the screening module 406 is further configured to divide the output value sample set into multiple output value sample subsets under multiple correlation levels based on the correlation coefficient between each output value sample in the output value sample set and the enterprise output value; for each output value sample subset under the correlation level, obtain multiple initial output value prediction models to be trained; train each initial output value prediction model to be trained based on the output value sample subset to obtain multiple trained initial output value prediction models; optimize the hyperparameters of each trained initial output value prediction model through the particle swarm optimization algorithm to obtain multiple candidate output value prediction models; and select the target output value prediction model of the correlation level from the multiple candidate output value prediction models based on the mean absolute percentage error of each candidate output value prediction model.

[0168] In one embodiment, the screening module 406 is further configured to calculate the correlation coefficient between each output value sample and the enterprise's output value; if the correlation coefficient is greater than a first preset correlation threshold, construct a first-level output value sample subset based on the corresponding output value sample; if the correlation coefficient is less than the first preset correlation threshold but greater than a second preset correlation threshold, construct a second-level output value sample subset based on the corresponding output value sample; if the correlation coefficient is less than the second preset correlation threshold but greater than a third preset correlation threshold, construct a third-level output value sample subset based on the corresponding output value sample; and if the correlation coefficient is less than the third preset correlation threshold, remove the corresponding output value sample from the output value sample set.

[0169] In one embodiment, the prediction module 408 is further configured to obtain a correlation coefficient calculation function; based on the output value prediction sample, obtain the correlation coefficient between the output value prediction sample and the enterprise output value by solving the correlation coefficient calculation function; and determine the correlation level corresponding to the output value prediction sample according to the relationship between the correlation coefficient of the output value prediction sample and the preset correlation threshold.

[0170] Each module in the aforementioned enterprise output forecasting device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0171] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 5As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores enterprise data, regional macroeconomic data, and historical electricity consumption data for the enterprise to be predicted. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for predicting enterprise output.

[0172] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0173] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0174] Obtain historical electricity consumption data of the enterprise to be predicted, perform clustering processing on the historical electricity consumption data, and obtain the electricity consumption pattern characteristics of the enterprise to be predicted.

[0175] Obtain enterprise data and regional macroeconomic data of the enterprises to be predicted, extract features from historical electricity consumption data, enterprise data and regional macroeconomic data to obtain electricity consumption-related features, macroeconomic features and time features, and construct an output value sample set based on electricity consumption pattern features, electricity consumption-related features, macroeconomic features and time features;

[0176] Based on the output value sample set, the target output value prediction model is selected from multiple candidate output value prediction models;

[0177] Obtain a sample of the output value forecast for the enterprise to be predicted. Based on the correlation coefficient between the output value forecast sample and the enterprise's output value, determine the corresponding correlation level. Input the output value forecast sample into the target output value forecast model corresponding to the correlation level, and output the enterprise output value forecast for the enterprise to be predicted.

[0178] In one embodiment, when the processor executes the computer program, it further implements the following steps: For each month, based on the historical electricity consumption data of the month, the point-by-point electricity consumption data of each production line is converted into a two-dimensional daily electricity consumption feature matrix with dates as rows and time nodes as columns; a preset number of daily electricity consumption features are selected in the two-dimensional daily electricity consumption feature matrix as initial cluster centers; for each initial cluster center, the distance between each daily electricity consumption feature and the initial cluster center is calculated, and an electricity consumption pattern cluster is constructed based on the distance; the target cluster center of the electricity consumption pattern cluster is calculated, and it is determined whether the target cluster center meets the aggregation convergence condition. If the target cluster center does not meet the aggregation convergence condition, the target cluster center is replaced with the initial cluster center, and the step of calculating the distance between each daily electricity consumption feature and the initial cluster center is returned to continue execution until the target cluster center meets the aggregation convergence condition; based on the multiple electricity consumption pattern clusters obtained when the target cluster center meets the aggregation convergence condition, the electricity consumption pattern features of multiple production lines in the month are determined.

[0179] In one embodiment, when the processor executes the computer program, it also performs the following steps: concatenating the electricity consumption pattern characteristics, electricity consumption-related characteristics, macroeconomic characteristics, and time characteristics of each month to obtain the output value characteristics of multiple months; and using the output value characteristics of each month as output value samples to construct an output value sample set.

[0180] In one embodiment, when the processor executes the computer program, it further performs the following steps: dividing the output value sample set into multiple output value sample subsets under multiple correlation levels based on the correlation coefficient between each output value sample in the output value sample set and the enterprise output value; for each output value sample subset under each correlation level, obtaining multiple initial output value prediction models to be trained; training each initial output value prediction model to be trained based on the output value sample subset to obtain multiple trained initial output value prediction models; optimizing the hyperparameters of each trained initial output value prediction model using a particle swarm optimization algorithm to obtain multiple candidate output value prediction models; and selecting the target output value prediction model of the correlation level from the multiple candidate output value prediction models based on the mean absolute percentage error of each candidate output value prediction model.

[0181] In one embodiment, when the processor executes the computer program, it further performs the following steps: calculating the correlation coefficient between each output value sample and the enterprise's output value; if the correlation coefficient is greater than a first preset correlation threshold, constructing a first-level subset of output value samples based on the corresponding output value samples; if the correlation coefficient is less than the first preset correlation threshold but greater than a second preset correlation threshold, constructing a second-level subset of output value samples based on the corresponding output value samples; if the correlation coefficient is less than the second preset correlation threshold but greater than a third preset correlation threshold, constructing a third-level subset of output value samples based on the corresponding output value samples; and if the correlation coefficient is less than the third preset correlation threshold, removing the corresponding output value sample from the output value sample set.

[0182] In one embodiment, when the processor executes the computer program, it further performs the following steps: obtaining a correlation coefficient calculation function; based on the output value prediction sample, obtaining the correlation coefficient between the output value prediction sample and the enterprise output value by solving the correlation coefficient calculation function; and determining the correlation level corresponding to the output value prediction sample according to the relationship between the correlation coefficient of the output value prediction sample and a preset correlation threshold.

[0183] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0184] Obtain historical electricity consumption data of the enterprise to be predicted, perform clustering processing on the historical electricity consumption data, and obtain the electricity consumption pattern characteristics of the enterprise to be predicted.

[0185] Obtain enterprise data and regional macroeconomic data of the enterprises to be predicted, extract features from historical electricity consumption data, enterprise data and regional macroeconomic data to obtain electricity consumption-related features, macroeconomic features and time features, and construct an output value sample set based on electricity consumption pattern features, electricity consumption-related features, macroeconomic features and time features;

[0186] Based on the output value sample set, the target output value prediction model is selected from multiple candidate output value prediction models;

[0187] Obtain a sample of the output value forecast for the enterprise to be predicted. Based on the correlation coefficient between the output value forecast sample and the enterprise's output value, determine the corresponding correlation level. Input the output value forecast sample into the target output value forecast model corresponding to the correlation level, and output the enterprise output value forecast for the enterprise to be predicted.

[0188] In one embodiment, when the computer program is executed by the processor, it further implements the following steps: For each month, based on the historical electricity consumption data of the month, the point-by-point electricity consumption data of each production line is converted into a two-dimensional daily electricity consumption feature matrix with dates as rows and time nodes as columns; a preset number of daily electricity consumption features are selected in the two-dimensional daily electricity consumption feature matrix as initial cluster centers; for each initial cluster center, the distance between each daily electricity consumption feature and the initial cluster center is calculated, and an electricity consumption pattern cluster is constructed based on the distance; the target cluster center of the electricity consumption pattern cluster is calculated, and it is determined whether the target cluster center meets the aggregation convergence condition. If the target cluster center does not meet the aggregation convergence condition, the target cluster center is replaced with the initial cluster center, and the step of calculating the distance between each daily electricity consumption feature and the initial cluster center is returned to continue execution until the target cluster center meets the aggregation convergence condition; based on the multiple electricity consumption pattern clusters obtained when the target cluster center meets the aggregation convergence condition, the electricity consumption pattern features of multiple production lines in the month are determined.

[0189] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: concatenating the electricity consumption pattern characteristics, electricity consumption-related characteristics, macroeconomic characteristics, and time characteristics of each month to obtain the output value characteristics of multiple months; and using the output value characteristics of each month as output value samples to construct an output value sample set.

[0190] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: dividing the output value sample set into multiple output value sample subsets under multiple correlation levels based on the correlation coefficient between each output value sample in the output value sample set and the enterprise output value; for each output value sample subset under each correlation level, obtaining multiple initial output value prediction models to be trained; training each initial output value prediction model to be trained based on the output value sample subset to obtain multiple trained initial output value prediction models; optimizing the hyperparameters of each trained initial output value prediction model using a particle swarm optimization algorithm to obtain multiple candidate output value prediction models; and selecting the target output value prediction model of the correlation level from the multiple candidate output value prediction models based on the mean absolute percentage error of each candidate output value prediction model.

[0191] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: calculating the correlation coefficient between each output value sample and the enterprise's output value; if the correlation coefficient is greater than a first preset correlation threshold, constructing a first-level subset of output value samples based on the corresponding output value samples; if the correlation coefficient is less than the first preset correlation threshold but greater than a second preset correlation threshold, constructing a second-level subset of output value samples based on the corresponding output value samples; if the correlation coefficient is less than the second preset correlation threshold but greater than a third preset correlation threshold, constructing a third-level subset of output value samples based on the corresponding output value samples; and if the correlation coefficient is less than the third preset correlation threshold, removing the corresponding output value sample from the output value sample set.

[0192] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: obtaining a correlation coefficient calculation function; based on the output value prediction sample, obtaining the correlation coefficient between the output value prediction sample and the enterprise output value by solving the correlation coefficient calculation function; and determining the correlation level corresponding to the output value prediction sample according to the relationship between the correlation coefficient of the output value prediction sample and a preset correlation threshold.

[0193] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0194] Obtain historical electricity consumption data of the enterprise to be predicted, perform clustering processing on the historical electricity consumption data, and obtain the electricity consumption pattern characteristics of the enterprise to be predicted.

[0195] Obtain enterprise data and regional macroeconomic data of the enterprises to be predicted, extract features from historical electricity consumption data, enterprise data and regional macroeconomic data to obtain electricity consumption-related features, macroeconomic features and time features, and construct an output value sample set based on electricity consumption pattern features, electricity consumption-related features, macroeconomic features and time features;

[0196] Based on the output value sample set, the target output value prediction model is selected from multiple candidate output value prediction models;

[0197] Obtain a sample of the output value forecast for the enterprise to be predicted. Based on the correlation coefficient between the output value forecast sample and the enterprise's output value, determine the corresponding correlation level. Input the output value forecast sample into the target output value forecast model corresponding to the correlation level, and output the enterprise output value forecast for the enterprise to be predicted.

[0198] In one embodiment, when the computer program is executed by the processor, it further implements the following steps: For each month, based on the historical electricity consumption data of the month, the point-by-point electricity consumption data of each production line is converted into a two-dimensional daily electricity consumption feature matrix with dates as rows and time nodes as columns; a preset number of daily electricity consumption features are selected in the two-dimensional daily electricity consumption feature matrix as initial cluster centers; for each initial cluster center, the distance between each daily electricity consumption feature and the initial cluster center is calculated, and an electricity consumption pattern cluster is constructed based on the distance; the target cluster center of the electricity consumption pattern cluster is calculated, and it is determined whether the target cluster center meets the aggregation convergence condition. If the target cluster center does not meet the aggregation convergence condition, the target cluster center is replaced with the initial cluster center, and the step of calculating the distance between each daily electricity consumption feature and the initial cluster center is returned to continue execution until the target cluster center meets the aggregation convergence condition; based on the multiple electricity consumption pattern clusters obtained when the target cluster center meets the aggregation convergence condition, the electricity consumption pattern features of multiple production lines in the month are determined.

[0199] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: concatenating the electricity consumption pattern characteristics, electricity consumption-related characteristics, macroeconomic characteristics, and time characteristics of each month to obtain the output value characteristics of multiple months; and using the output value characteristics of each month as output value samples to construct an output value sample set.

[0200] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: dividing the output value sample set into multiple output value sample subsets under multiple correlation levels based on the correlation coefficient between each output value sample in the output value sample set and the enterprise output value; for each output value sample subset under each correlation level, obtaining multiple initial output value prediction models to be trained; training each initial output value prediction model to be trained based on the output value sample subset to obtain multiple trained initial output value prediction models; optimizing the hyperparameters of each trained initial output value prediction model using a particle swarm optimization algorithm to obtain multiple candidate output value prediction models; and selecting the target output value prediction model of the correlation level from the multiple candidate output value prediction models based on the mean absolute percentage error of each candidate output value prediction model.

[0201] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: calculating the correlation coefficient between each output value sample and the enterprise's output value; if the correlation coefficient is greater than a first preset correlation threshold, constructing a first-level subset of output value samples based on the corresponding output value samples; if the correlation coefficient is less than the first preset correlation threshold but greater than a second preset correlation threshold, constructing a second-level subset of output value samples based on the corresponding output value samples; if the correlation coefficient is less than the second preset correlation threshold but greater than a third preset correlation threshold, constructing a third-level subset of output value samples based on the corresponding output value samples; and if the correlation coefficient is less than the third preset correlation threshold, removing the corresponding output value sample from the output value sample set.

[0202] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: obtaining a correlation coefficient calculation function; based on the output value prediction sample, obtaining the correlation coefficient between the output value prediction sample and the enterprise output value by solving the correlation coefficient calculation function; and determining the correlation level corresponding to the output value prediction sample according to the relationship between the correlation coefficient of the output value prediction sample and a preset correlation threshold.

[0203] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0204] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0205] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0206] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for predicting enterprise output value, characterized in that, The method includes: Obtain historical electricity consumption data of the enterprise to be predicted, and perform clustering processing on the historical electricity consumption data to obtain the electricity consumption pattern characteristics of the enterprise to be predicted. Obtain enterprise data and regional macroeconomic data of the enterprise to be predicted, extract features from the historical electricity consumption data, enterprise data and regional macroeconomic data to obtain electricity consumption-related features, macroeconomic features and time features, and construct an output value sample set based on the electricity consumption pattern features, electricity consumption-related features, macroeconomic features and time features; Based on the output value sample set, a target output value prediction model is selected from multiple candidate output value prediction models; Obtain the output value prediction sample of the enterprise to be predicted, determine the corresponding correlation level based on the correlation coefficient between the output value prediction sample and the enterprise's output value, input the output value prediction sample into the target output value prediction model corresponding to the correlation level, and output the enterprise output value prediction value of the enterprise to be predicted.

2. The method according to claim 1, characterized in that, The historical electricity consumption data includes point-by-point electricity consumption data for multiple production lines across multiple months; the clustering process of the historical electricity consumption data to obtain the electricity consumption pattern characteristics of the enterprise to be predicted includes: For each month, based on the historical electricity consumption data of that month, the point-by-point electricity consumption data of each production line is converted into a two-dimensional daily electricity consumption feature matrix with date as the row and time node as the column; A predetermined number of daily electricity consumption features are selected as initial cluster centers from the two-dimensional daily electricity consumption feature matrix; For each initial cluster center, calculate the distance between each daily electricity consumption feature and the initial cluster center, and construct an electricity consumption pattern cluster based on the distance; Calculate the target cluster center of the electricity consumption pattern cluster, determine whether the target cluster center meets the aggregation convergence condition, if the target cluster center does not meet the aggregation convergence condition, replace the target cluster center with the initial cluster center, and return to the step of calculating the distance between each daily electricity consumption feature and the initial cluster center to continue execution until the target cluster center meets the aggregation convergence condition; Based on the multiple electricity consumption pattern clusters obtained when the target cluster center satisfies the aggregation convergence condition, the electricity consumption pattern characteristics of multiple production lines in the month are determined.

3. The method according to claim 1, characterized in that, The process of constructing a sample set of output value based on the electricity consumption pattern characteristics, the electricity consumption-related characteristics, the macroeconomic characteristics, and the time characteristics includes: By concatenating the electricity consumption pattern characteristics, electricity consumption-related characteristics, macroeconomic characteristics, and time characteristics of each month, the output value characteristics of multiple months are obtained. The output characteristics of each month are used as output samples to construct an output sample set.

4. The method according to claim 1, characterized in that, The step of selecting a target output prediction model from multiple candidate output prediction models based on the output sample set includes: Based on the correlation coefficient between each output value sample and the enterprise output value in the output value sample set, the output value sample set is divided into multiple output value sample subsets under different correlation levels. For each relevant level's output value sample subset, multiple initial output value prediction models to be trained are obtained. Based on the output value sample subset, each initial output value prediction model to be trained is trained to obtain multiple trained initial output value prediction models. Through particle swarm optimization algorithm, hyperparameters of each trained initial output value prediction model are optimized to obtain multiple candidate output value prediction models. Based on the mean absolute percentage error of each candidate output value prediction model, the target output value prediction model for the relevant level is selected from the multiple candidate output value prediction models.

5. The method according to claim 4, characterized in that, The correlation levels include a first correlation level, a second correlation level, and a third correlation level; the step of dividing the output value sample set into multiple output value sample subsets under multiple correlation levels based on the correlation coefficient between each output value sample and the output value of the enterprise to be predicted includes: Calculate the correlation coefficient between each output value sample and the output value of the enterprise to be predicted; If the correlation coefficient is greater than the first preset correlation threshold, a subset of output value samples of the first correlation level is constructed based on the corresponding output value samples. If the correlation coefficient is less than the first preset correlation threshold and greater than the second preset correlation threshold, a subset of output value samples of the second correlation level is constructed based on the corresponding output value samples. If the correlation coefficient is less than the second preset correlation threshold and greater than the third preset correlation threshold, a subset of output value samples of the third correlation level is constructed based on the corresponding output value samples. If the correlation coefficient is less than the third preset correlation threshold, the corresponding output value sample is removed from the output value sample set.

6. The method according to claim 1, characterized in that, The step of determining the corresponding correlation level based on the correlation coefficient between the output value forecast sample and the enterprise output value includes: Obtain the correlation coefficient calculation function; Based on the output value forecast sample, the correlation coefficient between the output value forecast sample and the enterprise output value is obtained by solving the correlation coefficient calculation function; Based on the relationship between the correlation coefficient of the output value prediction sample and the preset correlation threshold, the correlation level corresponding to the output value prediction sample is determined.

7. A device for predicting enterprise output value, characterized in that, The device includes: The clustering module is used to obtain historical electricity consumption data of the enterprise to be predicted, and to perform clustering processing on the historical electricity consumption data to obtain the electricity consumption pattern characteristics of the enterprise to be predicted. The construction module is used to acquire enterprise data and regional macroeconomic data of the enterprise to be predicted, extract features from the historical electricity consumption data, the enterprise data and the regional macroeconomic data to obtain electricity consumption-related features, macroeconomic features and time features, and construct an output value sample set based on the electricity consumption pattern features, the electricity consumption-related features, the macroeconomic features and the time features. The filtering module is used to filter out the target output prediction model from multiple candidate output prediction models based on the output sample set; The prediction module is used to obtain the output value prediction sample of the enterprise to be predicted, determine the corresponding correlation level based on the correlation coefficient between the output value prediction sample and the enterprise's output value, input the output value prediction sample into the target output value prediction model corresponding to the correlation level, and output the enterprise output value prediction value of the enterprise to be predicted.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images